The University of Washington Physicians Network (UWPN) is a nonprofit primary care delivery system in Seattle with nine clinics linked by a computerized, networked clinical information system. In 1998, UWPN initiated a diabetes management program in collaboration with the Institute for Healthcare Improvement. In the process of developing the diabetes program, several successive types of registry systems were constructed using a computerized spreadsheet and a relational database. This project has clarified some of the challenges that any delivery system seeking to construct a diabetes management program must face regardless of whether they have paper charts or a completely computerized medical record system. As more providers make the transition to computerized clinical information systems, there will be greater opportunities to use the information residing in electronic charts for population-based care of chronic diseases like diabetes.

This article reviews the conceptual framework for chronic illness management programs. It then addresses specific key issues pertaining to construction of a registry for diabetic patients using readily available software.

The Conceptual Framework for Chronic Illness Management
Chronic illnesses such as diabetes are characterized by gradually worsening symptoms. Deterioration in both functional status and overall health can be minimized by optimal care, most of which is dependent on choices made by patients on a daily basis, usually involving little or no consultation with medical professionals.

Research has identified four essential tasks that people with chronic illnesses must perform if they are to keep their risk of excess morbidity to a minimum.¹ They must

1. engage in activities that promote health and build physiological reserve, including exercise, nutrition, social activities, and sleep.
2. interact with health care providers and systems and understand and adhere to recommended treatment protocols.
3. monitor their own physical and emotional status and make appropriate management decisions on the basis of symptoms and signs.
4. manage the impact of their illness on their own ability to function in important roles, on their emotions and self-esteem, and on their relationships with others.

Likewise, there are three essential activities that health care systems must carry out for patients with chronic illness. They must

1. assure delivery of those interventions (evaluations and treatments, both medical and psychosocial) that have been shown by rigorous evidence to be effective.
2. empower patients to take responsibility for the management of their conditions.
3. provide information, support, and resources to assist patients in self-management tasks.

Although the above tasks seem relatively simple, some of them are often overlooked in the course of scheduled office visits. Patients and delivery systems often fall short of achieving optimal clinical outcomes.^2,3

Following are reasons why this is the case, despite the fact that physicians and other members of the health care system strive to provide the best care possible.

1. Chronic illness care involves tasks that must be repeated over long intervals of time.
2. The amount of time primary care physicians are allotted per patient is usually 15–20 min, with pressures to see more patients, most of whom have either some acute problem or a mix of acute and chronic conditions.
3. It is often difficult for physicians to identify which patients with chronic illness they are responsible for treating. Patients who fail to schedule follow-up appointments may be overlooked by even the most attentive physicians.
4. Physicians are under tremendous pressure to not miss serious disease. As a result, visits are structured to specifically scan for symptoms, and it is easy for a busy physician to overlook routine monitoring tests such as HbA_1c or screening for microalbumin. All clinicians struggle to balance the needs of the chronically ill with the fast pace of a busy office practice.

The evidence for the effectiveness of chronic illness programs is based in successful clinical trials. However, such clinical trials have several organizational characteristics that set them apart from the usual office practice environment in which busy clinicians attempt to achieve outcomes similar to those described in the trials.⁴

1. In most large clinical trials, non-physician team members, including nurses and medical assistants, carry out the day-to-day tasks of the intervention.
2. Large clinical trials use computerized registries to keep track of all the patients entering the trial who enroll at different times over many months.
3. Large clinical trials use protocol-based interventions with automated reminders for when interventions are due.

Wagner and associates have developed a model for chronic illness management in clinical practice based on the system characteristics shown to improve outcomes in a number of chronic illnesses (Fig. 1).⁵ The model postulates the crucial dynamic in care delivery to be an informed and activated patient interacting with a prepared and proactive practice team to support self-management and behavior change. The basic categories of available resources and tools are 1) community services and policies, 2) decision support, 3) delivery system design, and 4) clinical information systems.

Community services and policies include public and private agencies outside the health care delivery system offering resources for patients and their families.

Decision support includes evidence-based guidelines and point-of-care reminders to minimize variation in the clinical strategies used throughout a delivery system in the management of patients with a chronic illness.⁶

Delivery system design refers to changes in processes of care delivery that help providers carry out the tasks of chronic illness care amid the competing demands of a busy office practice.

There are five types of delivery system design changes that have been shown to improve chronic illness outcomes. They are categorized as follows:

1. Nonphysician case managers assure that all chronic illness interventions occur at the appropriate times.⁷
2. Patients attend group/cluster visits for education and self-management support.⁸
3. Patients attend yearly structured visits with the primary care provider to review the specific chronic illness.⁹
4. Patients attend yearly structured visits with a multidisciplinary team, including a specialist, the primary care provider, and ancillary services.
5. Patients attend structured, half-day, Patients attend structured, half-day, Patients attend structured, half-day, structured assessment for specific aspects of the illness.¹⁰

Clinical information systems refers to the organization of clinical data from patient medical records to coordinate the delivery of interventions and self-management support activities of the entire population. Creation of a disease registry is crucial in the management of a population of patients with chronic illness.¹¹

Further research is needed to determine the relative importance of each of the four components of the chronic illness model. However, to achieve meaningful improvement in outcomes, a program must include some type of registry to identify and track patients, at least one element of delivery system design change, and some type of decision support.^12,13

The Computerized Disease Registry for Diabetes
The remainder of this article will outline an approach to creating a disease registry and will cover strategies for identifying patients, the type of information to include, ways of managing the registry, and the advantages and disadvantages of different types of software programs used for disease management. Some of the tools described are appropriate for large practices, whereas others may be used by small individual practices including providers working in rural areas.¹⁴

Figure 1. The chronic illness model. Adapted from reference 5.

How Do You Identify Patients to Include?
Identifying patients with diabetes can be done several ways.

1. Patients can be chosen who have a problem list entry or a diabetes-related International Classification of Diseases, Ninth Revision (ICD-9) code as an encounter diagnosis (250.00–250.99). Use of the encounter diagnosis to identify patients with diabetes may identify patients for whom a 250 code was used when ordering a test to rule out diabetes and thus will include patients without diabetes. In an electronic medical record, the encounter diagnoses cannot be changed once an encounter is closed. The problem list is designed to have problems added or deleted and may be a better flag for diabetes, since providers have an incentive to keep problem lists on their patient charts current.
2. Patients can be identified using laboratory values meeting a criterion for diabetes. A fasting serum glucose >126 mg/dl or a random glucose >200 mg/dl would be consistent with American Diabetes Association diagnostic criteria.¹⁵ HbA_1c should not be used for case finding because of lack of sensitivity.¹⁶
3. Hypoglycemic medications are quite specific for diabetes. Depending on the information system being used, it may be possible to generate a list of patients likely to have diabetes on the basis of medications prescribed.

All of these methods will result in the inclusion of some patients who do not have diabetes and occasionally miss some patients who do. The most practical approach is to use the above methods to identify candidates for inclusion in the diabetes registry. Then, as a second step, someone with clinical expertise can place on the registry those patients who clearly have diabetes.

A registry must include a mechanism to distinguish between patients who have left a practice and those who have simply not come in for a visit. Most patients with diabetes will be seen at least 3 times in 12 months at a primary care office. Any patient who has not been in contact with the clinic in 12 months should be contacted to determine whether they are overdue for monitoring or whether they have left the practice.

This is important for two reasons: 1) provider acceptance of outcome data is directly related to its perceived reliability, and 2) if a change in the care process is implemented as a pilot project, it is essential to be able to determine whether that innovation results in an improvement. If the innovation increases the chance of detecting patients who have left the practice, the noninnovation group will appear to have worse outcomes because the denominator will include more patients who have actually left the practice but appear to be overdue for an intervention.

The registry should include all patients with diabetes for whom the program using the registry is responsible. Responsibility for diabetes crosses several specialty interfaces, the most important of which is the interface between primary care and endocrinology. There may be no reason to include in a primary care registry patients for whom all aspects of diabetes care are managed by endocrinology. In some cases, glycemic control may be followed by an endocrinologist whereas risk factors for cardiovascular disease, such as hypertension, lipids, and smoking cessation, are issues for which primary care is responsible. In this situation, data from specialty visits should be included in the primary care registry to avoid redundant testing and confusion over whether the primary care physician or the endocrinologist is responsible for a particular aspect of the patient's illness. Shared management arrangements require close collaboration between primary care and endocrinology.

What Information Goes in the Registry?
The registry needs to include both dates and values for crucial interventions so as to identify patients who are either overdue for a monitoring intervention (such as blood pressure, glycosylated hemoglobin, or lipids) or have a laboratory result showing the need for further therapeutic intervention. Data in the registry can be categorized as 1) demographics, 2) glycemic control, 3) coronary risk factors, and 4) microvascular end organ disease data. A logical way to lay out the registry is shown in Figure 2. Only one variable for each category has been included for the sake of simplicity.

Additional data fields, as outlined below, should be considered for each of the categories depending on the ease of data capture and utility of the data.

1. Demographics: age, sex, type of diabetes, date of diagnosis, payor, and zip code.
2. Glycemic control: HbA_1c results, results of home blood glucose monitoring, and medications.
3. Coronary risk factors: blood pressure, lipids (total cholesterol, high-density lipoprotein [HDL], low-density lipoprotein [LDL], triglycerides), smoking status, and medications.
4. Microvascular end organ disease data: microalbumin screening; monofilament test for neuropathy; dilated retinal exam; for those with albumin/creatinine ratio >30, a measure of renal function such as creatinine clearance; and medications.

Where Do You Get the Data?
There are three ways to enter data into a registry. Data can be

1. hand entered directly into a spreadsheet from a medical chart.
2. extracted from electronic databases (billing records, pharmacy, or lab data).
3. downloaded from data fields in an electronic medical record (EMR).

An EMR in normal operation is perpetually in motion and constantly changing. During the day, clinical data are added continuously, wheras at night, data are entered by results reporting and home access by providers on call.

Before clinical data from patient visits, laboratory results, and pharmacy can be used for a registry or any kind of reporting, they must be downloaded into a static form either as a spreadsheet or a relational database, as shown in Figure 3. For this to occur, a relational database must be built with defined data fields to contain each specific data type in the registry. One such data field, for example, would be date of last urine albumin/creatinine ratio, whereas on another would be the value for that test. Programmed instructions for downloading data are then written so that each selected data type in the electronic medical record will be routed to the appropriate data field.

Figure 3. A networked EMR and relational database.

Data downloads can be scheduled to occur nightly so that the data for any given report are never more than 24 h old. The interface for this kind of download can be a challenge to maintain as software programs are upgraded, laboratories make changes in their tests, and medications are added to or deleted from a formulary.

In delivery systems with less than full computerization, there are still ways to use electronic records in a registry. In small or individual medical practices, demographics such as age, sex, payor, and zip code are often obtainable from computerized billing records. Likewise, it may be possible to obtain laboratory data on specific patients that can be entered electronically into a registry.

On the other hand, even with a full EMR, clinical staff may enter data into the chart as text but not in a form that is accessible to a computer query. To be accessible, information relevant to disease management must reside in a data field. For example, if the results of monofilament screening for neuropathy are buried in a chart note, the only way they can be recovered is by chart review. A data entry window for results of neuropathy screening will enter the data into a specific field in the EMR, which can be downloaded to the registry and accessed by query.

A similar issue is encountered with blood pressure, which if entered in standard format in a data field as a string variable cannot be used to reliably identify patients with hypertension. The blood pressure must first be converted to two separate data fields—one for systolic and one for diastolic pressure—so that it can be formatted as a numeric variable. This allows a query to be written identifying patients with either systolic or diastolic hypertension.

Whose Job Is It?
Although it is important to automate as much of the data entry as possible, there will often be data that need to be entered by hand. In small individual practices or clinics with paper record systems, the task of identifying patients to add to the registry and keeping the registry data current can be done by one individual with clerical or entry-level clinical experience. In larger systems, it is often more efficient to designate a disease management person (a nurse certified diabetes educator or a mid-level provider with clinical expertise in diabetes) to enter data and oversee the workings of the registry. Although individual physicians will often enjoy working with data in the registry and should be encouraged to participate in multiple levels of disease management activities, it is seldom cost-effective to have physicians performing registry maintenance tasks.

How Will the Registry Be Used?
"What information do the primary care providers and their team need in order to manage their diabetic patients?" This is the most important question to answer before deciding how to structure the registry and what the reports will look like.

In general, providers need the names of patients who need specific things done within a specific time frame. For example, a delivery system may monitor quality of care based on the percentage of diabetic patients who have had an HbA_1c measured within 6 months. What providers then need is a list each month of the patients with diabetes who have not had an HbA_1c measured within 5 months. A team member can then call those patients with a reminder to be tested. In this way, the team has time to act on the information from the registry to improve the outcomes and meet the organizational goals.

Likewise, the organization may select a target of less than some percentage of their patients with diabetes having an HbA_1c >7, 8, or 9%. The providers need an accurate and timely list of patients whose most recent HbA_1c is greater than the target so they can be brought in for more intense intervention. The report can provide similar information for coronary risk factor reduction, such as lipids and blood pressure testing, or for end organ disease monitoring, including microalbumin screening. The report should be directly linked to explicit organizational goals for improved outcomes with clear tasks for the team to perform to achieve the goals.

The frequency of the reports will depend on their use and on how often the data in the registry are updated. For a diabetes management program run by a chronic illness nurse, weekly reports may be crucial, particularly if the reports can be automatically generated from a database that is updated nightly. In a fully automated EMR with a relational database, it may be possible to run the reports daily and post them on an Intranet URL, which anyone on the clinical team can access. On the other hand, if data are hand entered into a spreadsheet and reports take a significant amount of time to prepare, monthly or even quarterly reporting may suffice.

How Should the Registry Be Structured?
There are two desktop computing tools for registry design: a spreadsheet and a relational database. It is a good idea to be familiar with both. In general, spreadsheets are easier to use without acquiring extensive computer skills and are very accessible to small clinics and individual practices, including rural areas. A relational database has greater potential for automated reporting, making it often the most appropriate tool for large integrated group practice settings.

Spreadsheets
Advantages

1. Visibility. As shown in the simplified spreadsheet in Figure 2, the names of all of the patients and their data are visible simply by scrolling around the screen. The ability to see patients' names and their data may help physicians and other team members make the conceptual change to thinking of the population as a whole.
2. Ability to interact with the data. By sorting data in a spreadsheet, motivated providers may be stimulated to use the population-wide perspective to begin to ask clinical questions about their population of patients. A simple sort by HbA_1c date will tell a viewer the names of all the patients who are overdue for glycemic monitoring. On a more complex level, one might ask, "Who are my patients whose triglycerides are too high to see the LDL and who also have an HbA_1c that is adequately controlled and therefore should be started on a triglyceride-lowering agent?"
3. Simplicity. Many providers, nurses, and medical assistants have basic spreadsheet skills, and those who do not can often be trained quickly.
4. Ease of data entry. Data entry for spreadsheets is simple. Although data entry screens can be created with more advanced programming skills, data are usually entered directly into the spreadsheet.
5. Ease of data cleaning. Cleaning the data is facilitated by simple maneuvers such as sorting by column and splitting the screen. In this way, numbers that do not make sense usually stand out to the viewer.

Disadvantages

1. Difficulty of macros. The reporting tools within spreadsheets can be automated to generate reports, but to do so requires an intermediate to high level of programming skill. Generating reports without macros is time consuming and requires repeated sorting and careful tracking of numerators and denominators.
2. Labor-/time-intensiveness. The number of patients who can be followed using a hand-maintained spreadsheet run by one person is probably limited to <1,000. The complexity of the process usually becomes the motivation for finding ways to automate.
3. Unwieldiness when used for multiple disease registries. Once a delivery system develops multiple chronic disease management programs, a decision must be made regarding whether to have a single spreadsheet for all diseases regardless of their degree of overlap or whether to develop multiple registries for different diseases.

   Use of a single spreadsheet for all diseases quickly results in a huge data file that   can be cumbersome to manage. Use of different registries for separate diseases will result in individuals with multiple comorbidities requiring duplicate data entry into multiple spreadsheets.

Relational Databases
Relational databases were designed to deal with situations in which data must be used in multiple ways depending on the context in which they are needed. A relational database allows users to design different data tables for each aspect of care, all of which are linked by unique identifiers—in this case, the patient record number. One table will contain all demographic information pertaining to a patient, another table will contain the clinic and primary care provider, a third table will contain clinical information for visits, such as provider, vital signs, and so forth. Laboratory data will reside in a fourth table and medications in a fifth. Queries are written in programming language to take appropriate data from each table to be included in a report. The reports can be written to include names of patients as well as summary statistics.

Advantages

1. Automation. A limited number of questions need to be routinely asked in the management of diabetes. Those questions can be used to build automated queries and reports, which can be run from the database at predetermined intervals and delivered to the appropriate people. The basic questions are, "Who are the patients with diabetes for whom we are responsible who have not had the following tests within the agreed-upon time frame" and "Whose results were worse than the agreed-upon target?":

   HbA1c Blood LDL Microalbumin Monofilament screening for neuropathy Dilated retinal exam

2. Roll-up/drill-down features. The same reports providing the names of patients and clinical outcomes for individual practices can be aggregated to provide outcomes data for a clinic or for the entire delivery system for quality improvement reporting.
3. Facility of multiple disease management programs. If a delivery system develops additional disease management programs, such as asthma or heart failure, a patient with multiple illnesses can be followed in each program without duplication of data entry.

Disadvantages

1. Complexity. Building the relational database and the query/report system require informatics expertise. Disease management programs require computer programmers and other nformation system personnel.
2. Difficulty of error identification. Often the only way to determine whether an error lies within the data or in the programming language of the report is for a clinician who knows the patients to whom the data pertain and a programmer who understands the program to work closely together.
3. Expense of data entry tools. Although a relational database can be set up with data entry screens so that the data are hand-entered, this is a complex task. Use of a relational database for disease management is most appropriate for delivery systems in which there exists a high degree of computerization so that data are automatically entered in the relational database from the EMR.
4. System instability. Once automated queries are running properly, any change in the data from which the database is derived, such as a change in the type of test done for HbA_1c which changes the identifier for the test result, can cause the entire system to stop working.

Conclusion
Efforts to achieve improved outcomes for diabetes require an organized population-based approach to diabetes management using all of the components of the chronic illness care model. Insurance payors are beginning to contract with commercial disease management companies on a national scale in an effort to control the high cost of diabetes care. Any delivery system seeking to manage their patients with diabetes using a disease management program that is internally integrated within their primary care delivery system must develop a registry for diabetes as an essential element of the program.

For beginning programs, it makes sense to start with a spreadsheet. Pilot projects, demonstration projects, and other small-scale efforts to develop disease management experience will often be best served by using a spreadsheet that is maintained by a person intimately involved in the use of the data the spreadsheet contains.

The tasks associated with developing a relational database, if begun at the start of a chronic illness program, are complex enough that efforts to make the information system work properly may impede other aspects of the program. If a small-scale project is successful enough to continue and spread to the entire organization, it will probably be necessary to convert to a relational database. A disease management program that is mature enough to function out of several clinics and has sufficient organizational support to have allocated to it dedicated informatics resources will find a registry using a relational database much more efficient.

Jeffrey Hummel, MD, MPH, is Director of Research and Clinical Improvement at the University of Washington Physicians Network in Seattle.